This paper describes an approach that uses support vector machines (SVM) for path planning of mobile robots. The algorithm generates a collision free path for mobile robots running between two tracks or moving towards a known way point. This approach can negotiate tracks and avoid obstacles which may be initially unknown but are later perceived by the robot, and hence is suitable for use with onboard sensors which provides local information. The approach involves dividing the whole terrain into two different classes, classifying any new point obtained from sensors into either of the classes, and generating a track between both the classes as a path of the robot. SVM generates a non-linear class boundary on the principle of maximizing the margin. The boundary generated by this method is smooth, free of obstacles, and safe for a robot to navigate. The paper presents various case studies and simulation results. Future possibility to integrate this technique with other path planning techniques is also discussed.
Graph Neural Networks for Decentralized Multi-Robot Path Planning
